Preserving Privacy in Medical Data: Homomorphic Encryption-Enhanced Collaborative ML

Abstract

Abstract Healthcare industry is going from massive transformation motivated by advancements in tools and technology and various data driven approaches. In this healthcare data represent various challenges like privacy, interoperability, and security. As the healthcare sector embraces technological advances, concerns about the privacy and security of critical patient data have become more prevalent. Collaborative machine learning (CML) and Homomorphic Encryption (HE) have become viable approaches to these challenges that allow useful data exchange and analysis. This research project discusses the cryptographic techniques that allow computations on encrypted data while delving into the fundamental ideas of HE. Simultaneously, it explores various frameworks for CML and highlights their potential for decentralized model training. The research project also critically analyzes the benefits and challenges of integrating HE with CML, offering insights into safe model aggregation, guaranteeing data privacy, and performance optimization techniques for use in healthcare environments. This research project delves into pragmatic scenarios and actual implementations, illuminating the ways in which the unified framework might improve diagnosis and cooperative research endeavors. This research project evaluates how well k-nearest neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) perform in a particular case study. The results showed that KNN had the best accuracy of 76.5%, with RF and SVM both had the accuracy of 76%. LR performed somewhat worse i.e. 73.5%. These findings offer insightful information for selecting models that take accuracy and the trade-off between precision, recall, and F1 score into account. This helps researchers make well-informed selections for their classification assignments. In this research project, we also have developed a privacy preserving framework using homomorphic encryption and collaborative machine learning. And to do the thorough investigation of the framework, we have applied various machine learning algorithm like Random Forest, Neural Network, Logistic Regression, Naïve Bayes, k-Nearest Neighbors and Support Vector Machine on different encryption techniques like Symmetric Encryption, Asymmetric Encryption, Hash Functions and Homomorphic Encryption. Each encryption technique has its own advantages and limitations which are discussed in detail in this research project. Also, this research project tries to demonstrate the effectiveness of Homomorphic encryption in real time applications where privacy preservation is a big challenge. Our study also adds to the existing conversation on improving healthcare analytics in a privacy-centric framework by bridging the gap between privacy preservation and data-driven insights.